Min-Sum Decoding

How min-sum approximates belief propagation and why normalisation, offset and quantisation matter.

Min-sum is a common approximation to the sum-product check-node update used in LDPC decoding. It replaces a more expensive nonlinear calculation with sign operations and a minimum-magnitude rule, which makes the decoder attractive for hardware.

Core idea

A basic min-sum check node computes the product of incoming signs and uses the smallest incoming magnitude, excluding the destination edge. Because plain min-sum tends to overestimate message magnitude, practical decoders often use normalised min-sum or offset min-sum to reduce that bias.

Engineering relevance

The best scaling or offset depends on the code, quantisation and channel model. A value tuned for floating-point simulation may not remain optimal after fixed-point clipping and saturation are introduced. Hardware design must also handle first/second minimum tracking efficiently for parallel check updates.

Key points

  • Min-sum simplifies the check-node computation.
  • Normalisation or offset can improve the approximation.
  • Fixed-point quantisation can change the optimum tuning.
  • First/second minimum logic is central to efficient hardware implementations.
Public technical resource

This page explains established concepts and engineering principles. Proprietary JobMettle methods and internal ThinkQuick technology are not disclosed.